Report #74226
[architecture] Treating all memory as flat text chunks making it impossible to distinguish facts from events
Partition memory into Episodic \(timestamped events\) and Semantic \(facts about the codebase\) stores. Query them differently based on the task.
Journey Context:
A flat vector store mixes facts with events. If the user asks why a decision was made, semantic search fails because the reason is buried in a past conversation. By separating them, the agent can query Episodic memory for 'decisions about database' and Semantic for 'current database config,' yielding precise answers for both temporal and structural queries.
⚠ Workarounds are unverified - always check before running. Confirmations show what worked for others, not a safety guarantee.
Lifecycle
2026-06-21T07:11:13.995348+00:00— report_created — created